Supporting Feature Extraction from Natural Language Requirements Specifications for Software Product Line
○肖 越東,久住憲嗣,福田 晃(九大)
Analyzing and extracting features from requirement specifications is an indispensable activity to support Software Product Line (SPL). However, performing features extraction is a time-consuming and inefficient task, since massive textual requirements need to be analyzed and classified. In this paper, we suggest employing some natural language processing approaches to support feature extraction. In particular, our technique consists of four steps: First, we identify a ranked list of terms from a requirement specification by two Part of Speech (POS) patterns. Second, we adopt an approach named contrastive analysis to identify domain-specific terms. Third, we utilize a pre-trained word2vec model to vectorize those terms. Finally, we feed the term vector matrix into Hierarchical Agglomerative Clustering algorithm to cluster features. Initial results show that accuracy is still limited, but that our approach allows automating the entire process.

footer 著作権について 倫理綱領 プライバシーポリシー セキュリティ 情報処理学会